Intrinsic Knowledge Evaluation on Chinese Language Models

TL;DR

This paper introduces a comprehensive Chinese language model knowledge evaluation benchmark covering syntactic, semantic, commonsense, and factual knowledge, providing insights into what and how well models encode various knowledge types.

Contribution

It presents four new evaluation tasks and a large question dataset to assess Chinese language models' encoding of different knowledge aspects, filling a gap in current evaluation methods.

Findings

Proposed a reliable benchmark for Chinese LM knowledge evaluation

Demonstrated models' strengths and weaknesses across knowledge types

Provided publicly available dataset for future research

Abstract

Recent NLP tasks have benefited a lot from pre-trained language models (LM) since they are able to encode knowledge of various aspects. However, current LM evaluations focus on downstream performance, hence lack to comprehensively inspect in which aspect and to what extent have they encoded knowledge. This paper addresses both queries by proposing four tasks on syntactic, semantic, commonsense, and factual knowledge, aggregating to a total of $39,308$ questions covering both linguistic and world knowledge in Chinese. Throughout experiments, our probes and knowledge data prove to be a reliable benchmark for evaluating pre-trained Chinese LMs. Our work is publicly available at https://github.com/ZhiruoWang/ChnEval.